minimum number of observations - document or check ?

[josef-pkt]

grangercausalitytests uses OLS for estimation. It breaks, raises exception at different (random looking) points if the number of observations is too small.

from a quick calculation, it looks like the minimum number of observations is

nobs > 3 * maxlag + addconstant

see mailing list thread 2012-06-29: "Strange error in statsmodel​s.tsa.stat​tools.gran​gercausali​tytests"

document or check and raise ValueError, when we know what the minimum number of observations is (problem perfect multicollinearity, singular design)

[josef-pkt]

http://stackoverflow.com/questions/11265518/adf-test-in-statsmodels-in-python

>>> [int(np.ceil(12. * np.power(nobs/100., 1/4.))) for nobs in np.arange(15)]
[0, 4, 5, 5, 6, 6, 6, 7, 7, 7, 7, 7, 8, 8, 8]
>>> range(15)
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14]

we need a different rule and a warning for small numbers

[josef-pkt]

move to 0.6 milestone (or it can maybe be done for 0.5.1)